Vision-threatening diabetic retinopathy (VTDR) is a severe complication of type 2 diabetes mellitus (T2DM) that can progress toward irreversible vision loss without timely detection and treatment. Early identification remains challenging because clinical evaluation often relies on subjective assessments, and validated biomarkers are limited. A study published in Frontiers in Endocrinology evaluated whether machine learning models could identify VTDR among individuals with diabetic retinopathy (DR).
The retrospective analysis used clinical data extracted from electronic medical records of patients with T2DM and DR treated at a hospital. Participants were categorized into VTDR and non-VTDR groups, with non-VTDR defined as mild-to-moderate non-proliferative diabetic retinopathy. The dataset was divided into training and testing cohorts using a 7:3 split. A total of eight machine learning models were developed and evaluated using metrics including area under the receiver operating characteristic curve (AUC), accuracy, precision, recall, Brier score, and F1 score. Model performance was further assessed using a comprehensive scoring system with a maximum score of 64, and Shapley Additive Explanations (SHAP) were applied to interpret the best-performing model.
Among 1,124 enrolled patients, the prevalence of VTDR was 36.9%. Factors observed in relation to VTDR included diabetic treatment, duration of T2DM, glycated hemoglobin levels, albuminuria, and anemia. The support vector machine (SVM) model demonstrated the best performance, achieving AUC 0.879, accuracy 0.837, precision 0.833, F1 score 0.756, and Brier score 0.129, and obtained the highest total score of 57 out of 64 in the testing cohort. Decision curve analysis and calibration curves supported the model’s performance, and a simplified calculator based on SHAP feature importance maintained diagnostic capacity.
These findings demonstrate the potential application of an interpretable machine learning approach for identifying VTDR among individuals with diabetic retinopathy.
